Raison du choix de MEP:

“Je serais plutôt pour tester un de ceux qui sort dans le papier d’Arianne, au choix : MEP, MBzP ou MnBP. Prends en un au hasard.” e-mail Claire 23/11/20

Since last time:

added exp time, child age, maternal depression

To do:

explore the possibility of exposure as long format (cf the paper claire reviewed)

1 Intro

On 10/20/2020 we discussed with CP of babylab’s analysis plan (cf meeting notes in project_log.html and model_choice.html. We still had questions regarding multiple exposures in the same model (T2 and T3) and regarding interactions between timing of exposure (T2 and T3) and timing of eyetracker experiment (M5, M12, M24). To finalise our analysis plan we decided to perform the following analysis:

model for one compound (MEP) at 24 months against mean fixation duration
do a model with T2, then T3 then T2 + T3
repeat for M5 and M12
do a mixed model with M24, M12 and M5
compare all these results to see if they are coherent and if it makes sense to have a repeated measure model with both exposure measurements
then add an interaction in the repeated measures model to see if different exposure times have different effects at different ages

2 Descriptive

2.1 MEP

=> MEP looks log noramal(-ish), will include as log in the model

2.2 Mean fixation duration

Arithmetic mean of “mean fixation duration” at face recognition and scene exploration tasks:

clear difference with age

better log transformed (may want to at least do sensitivity analyses on the log transformed)

2.3 Covariates

Selected covariates and coding see dag_var_coding.html:

maternal education
child sex
parity
tobacco
maternal age
maternal depression
child age
experiment time

3 Model with no exposure

At 24m

## 
## Call:
## lm(formula = fix_dur ~ mo_edu + ch_sex + mo_par + mo_tob + mo_age + 
##     mo_had_cat + exp_t_cat, data = model_data_24)
## 
## Residuals:
##     Min      1Q  Median      3Q     Max 
## -94.343 -29.934  -8.229  30.164 159.184 
## 
## Coefficients:
##                   Estimate Std. Error t value Pr(>|t|)    
## (Intercept)       260.3684    35.2128   7.394 1.55e-11 ***
## mo_edu<=bac+2      -7.9042    12.2693  -0.644  0.52056    
## ch_sex2            -4.3444     8.4277  -0.515  0.60708    
## mo_par1            -1.9774     9.0759  -0.218  0.82787    
## mo_par2             7.5433    17.3116   0.436  0.66375    
## mo_tob             34.3024    13.4164   2.557  0.01171 *  
## mo_age              3.1132     1.1129   2.797  0.00594 ** 
## mo_had_cat(8,11]  -12.4455    10.2704  -1.212  0.22779    
## mo_had_cat(11,25]   7.4022    10.2970   0.719  0.47351    
## exp_t_cat(10,11]  -22.5877    14.6573  -1.541  0.12573    
## exp_t_cat(11,14]    5.5768    16.5712   0.337  0.73701    
## exp_t_cat(14,16]   -0.1187    12.6825  -0.009  0.99255    
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## Residual standard error: 48.83 on 130 degrees of freedom
##   (9 observations deleted due to missingness)
## Multiple R-squared:  0.1404, Adjusted R-squared:  0.06763 
## F-statistic:  1.93 on 11 and 130 DF,  p-value: 0.04103

effect of tobacco and maternal age
same results for log transformed fix_dur

4 Models with exp

For values and p vals check annex.

4.1 Compared models

Linear regression for each age group (24m, 12m and 5m):

fix_dur ~ ln(MEP_t2) + covariates
fix_dur ~ ln(MEP_t2) + covariates
fix_dur ~ ln(MEP_t2) + ln(MEP_t3) + covariates

Mixed effect model for all age groups combined (repeated measures):

fix_dur ~ ln(MEP_t2) + age_group + covariates + 1|ident
fix_dur ~ ln(MEP_t3) + age_group + covariates + 1|ident
fix_dur ~ ln(MEP_t2) + ln(MEP_t3) + age_group + covariates + 1|ident

Mixed effect model with interaction term to combine age groups (repeated measures) and take into account possible differences in effect between time of exposure and age of measurement:

fix_dur ~ ln(MEP_t2) + age_group + covariates + ln(MEP_T1) * age_group + 1|ident
fix_dur ~ ln(MEP_t3) + age_group + covariates + ln(MEP_T3) * age_group + 1|ident
fix_dur ~ ln(MEP_t2) + ln(MEP_t3) + age_group + covariates + ln(MEP_t2) * age_group + covariates + ln(MEP_t3) * age_group + 1|ident

4.2 Simple models

No collinearity issues in the T2 + T3 model (see model diagnostics, eg 5.2.4)
T2 + T3 model not totally consistent with the separate T2 and T3 models
- interpret these differences

=> decide on whether we want to include T2 and T3 in the same model

effect consistent over age group for the separated T2 and T3 models
- this would be the green flag for a mixed model effect without interaction

4.3 Mixed effect models

Mixed model estimates consistent with separate age group estimates with smaller ICs
- this would confirm that we might be interested in the gain of power provided by the mixed effect model

4.4 Model with interactions

Note that interaction terms are not projected for the age group, they are just the interaction term.

Results consistent with the mixed model
Interaction terms close to zero
- this would confirm the unnecessity of adding the interaction term

4.5 Conclusions

to discuss with claire

=> I would do separate mixed models for T2 and T3 (most power/sensitive, more interesting interpretation wise) with the following sensitivity analyses

T2 + T3
separated age groups
interaction

5 Annex

5.1 Model values

## # A tibble: 16 x 6
##    term        estimate conf.low conf.high model   age  
##    <chr>          <dbl>    <dbl>     <dbl> <chr>   <fct>
##  1 log(MEP_T1)    0.661    -7.49      8.81 T2      24m  
##  2 log(MEP_T3)    5.20     -3.01     13.4  T3      24m  
##  3 log(MEP_T3)    7.69     -2.74     18.1  T2 + T3 24m  
##  4 log(MEP_T1)   -3.93    -14.2       6.29 T2 + T3 24m  
##  5 log(MEP_T1)    8.38     -5.49     22.3  T2      12m  
##  6 log(MEP_T3)    7.37     -7.42     22.2  T3      12m  
##  7 log(MEP_T3)    3.17    -15.4      21.7  T2 + T3 12m  
##  8 log(MEP_T1)    6.59    -10.9      24.1  T2 + T3 12m  
##  9 log(MEP_T1)    6.96    -12.0      25.9  T2      5m   
## 10 log(MEP_T3)    3.15    -20.7      27.0  T3      5m   
## 11 log(MEP_T3)   -5.59    -39.1      28.0  T2 + T3 5m   
## 12 log(MEP_T1)   10.1     -16.8      36.9  T2 + T3 5m   
## 13 log(MEP_T1)    2.96     -4.06      9.98 T2      mixed
## 14 log(MEP_T3)    5.47     -1.82     12.8  T3      mixed
## 15 log(MEP_T3)    5.84     -3.49     15.2  T2 + T3 mixed
## 16 log(MEP_T1)   -0.475    -9.38      8.44 T2 + T3 mixed

MEP analysis

modelling strategy validation

M. Rolland

07/01/2021

1 Intro

2 Descriptive

2.1 MEP

2.2 Mean fixation duration

2.3 Covariates

3 Model with no exposure

4 Models with exp

4.1 Compared models

4.2 Simple models

4.3 Mixed effect models

4.4 Model with interactions

4.5 Conclusions

5 Annex

5.1 Model values

5.2 Model diagnostics

5.2.1 No exposure model diagnostic

5.2.2 Simple model 24m T2

5.2.3 Simple model 24m T3

5.2.4 Simple model 24m T2 + T3

5.2.5 Simple model 12m T2

5.2.6 Simple model 12m T3

5.2.7 Simple model 12m T2 + T3

5.2.8 Simple model 5m T2

5.2.9 Simple model 5m T3

5.2.10 Simple model 5m T2 + T3

5.2.11 Mixed model T2

5.2.12 Mixed model T3

5.2.13 Mixed model T2 + T3

5.2.14 Mixed model + interaction model T2

5.2.15 Mixed model + interaction model T3

5.2.16 Mixed model + interaction model T2 + T3